Category

Safety: RL environments and AI benchmarks

AI benchmarks
124
Shares entries with
4 categories

Snapshot 2026-10-11. How categories are built · Where the data comes from

In short

What are RL environments for safety?
A simulated workplace where an AI agent practises a task (a code repository, a browser, business software) and an automatic grader scores each attempt. Labs train on thousands of these attempts. Bought to improve models.

Only benchmarks use the Safety tag: the RL environment directories file their vendors under other categories. Browse all RL environments.

AI benchmarks 124

None confirmed yet.

124 not yet confirmedListed in the catalog; no line from its own site or paper confirms it yet.
  1. Gray Swan IPI — Combined Attack Set
  2. AIMS
  3. CheckMIABench
  4. HarmVideoBench
  5. IndoBias
  6. MT-JailBench
  7. RealityTest
  8. RefusalBench
  9. StereoTales
  10. Triaging Threats to Specialized Guardrails
  11. InvisibleBench
  12. COMPL-AI
  13. LiveSecBench
  14. AI-Secure LLM Trustworthy Leaderboard
  15. PandaBench
  16. MIR-SafetyBench
  17. PersistBench
  18. RedactionBench
  19. SaLAD
  20. UbuntuGuard
  21. Plant
  22. ASPI
  23. DecodingTrust-Agent Platform (DTap)
  24. EUDAIMONIA: Evaluating Undesirable Dynamics in AI
  25. Every Act Has Its Price
  26. FBHM
  27. FORTIS: Benchmarking Over-Privilege in Agent Skills
  28. FraudSMSWalker
  29. Governance Decay
  30. HarnessAudit-Bench
  31. KSAFE-MM
  32. LITMUS
  33. LivePI
  34. LLM-as-a-Reviewer
  35. MAAT: Multi-phase Adapter-Aware Targeted Unlearning
  36. MIRAGE
  37. ML-Bench
  38. MonitoringBench
  39. MosaicLeaks
  40. MultiBreak
  41. MuPHI
  42. MuPPET
  43. OverEager
  44. PhoneSafety
  45. PluRule
  46. PreScam
  47. PseudoBench
  48. RedVox
  49. Relevance as a Vulnerability
  50. Reward Hacking Benchmark
  51. ROK-FORTRESS
  52. RuleSafe-VL
  53. SafeClawBench
  54. SciIntegrity-Bench
  55. SkillSafetyBench
  56. StrongREJECT
  57. StructBreak
  58. The Age of Curiosity Meets the Age of AI
  59. The Compliance Trap
  60. ThreatCore
  61. TukaBench
  62. When Context Flips
  63. When Lower Privileges Suffice
  64. XL-SafetyBench
  65. AgentLeak
  66. BioSecBench-Function
  67. BrowserART
  68. LLM-AggreFact
  69. HarmActionsEval
  70. BioSecBench-Refusal
  71. BioSecBench-Surveillance
  72. Halluverse-M3
  73. ThaiSafetyBench
  74. X-Risks Leaderboard
  75. Are LLM Decisions Faithful to Verbal Confidence?
  76. AttaQ
  77. CodeGuard: Improving LLM Guardrails in CS Education
  78. Contextual StereoSet
  79. CSSBench
  80. HoloFair
  81. MPCI-Bench
  82. OI-Bench
  83. Overlooked Safety Vulnerability in LLMs
  84. PluriHarms
  85. RedBench
  86. RubRIX
  87. VERA-MH
  88. When LLMs Imagine People
  89. XSTest
  90. Enhancing Agent Safety Judgment
  91. HARP
  92. Beyond Fixed Benchmarks and Worst-Case Attacks
  93. Decomposing and Measuring Evaluation Awareness
  94. EditRisk-Bench
  95. FORTRESS
  96. How Far Are VLMs from Privacy Awareness in the Physical World? An Empirical Study
  97. Measuring Safety Alignment Effects in Autonomous Security Agents
  98. MultiTurnPSB
  99. NRT-Bench
  100. PoisonForge
  101. Red Queen
  102. Scammer4U
  103. SomaliBench Eval
  104. Towards Understanding and Measuring COGNITIVE ATROPHY in LLM Behaviour
  105. Vera-Bench
  106. CAIS Risk Index
  107. SafeArena
  108. TOFU LLaMA 10%
  109. TOFU LLaMA 1%
  110. TOFU LLaMA 5%
  111. TOFU Phi 10%
  112. TOFU Phi 1%
  113. TOFU Phi 5%
  114. EvasionBench
  115. ChemSafetyBench
  116. PII Masking Benchmark
  117. NodeSynth: Socially Aligned Synthetic Data for AI Evaluation
  118. (implicit — evaluation of prompting-based defenses against domain-camouflaged injection)
  119. Integrated Safety Evaluation
  120. MemLeak
  121. The Heterogeneous Safety Impacts of Benign Multilingual Fine-Tuning
  122. The Misattribution Gap
  123. Unintended Long-Term State Poisoning Bench
  124. VisualLeakBench

Domains that overlap with Safety

Next: pick another domain. Tags come from the RL Research daily scrape, canonicalized nightly.